fetch/AbortSignal: honor init.signal + modernize AbortSignal — hop 4 of #196

[bkaradzic-microsoft]

Hop 4 of #196 — make AbortController / init.signal actually work in fetch(), and modernize the AbortSignal polyfill.

Problem

fetch() passed arcana::cancellation::none() for both continuations and never looked at init.signal, so abort never rejected (AbortError) and never cancelled the transport. The AbortSignal polyfill also predated the modern spec — no reason, no throwIfAborted(), a writable aborted, and no static AbortSignal.abort(). Combined with AbortController being installed globally in the Playground (BabylonJS/BabylonNative#1708), library feature-detection passed while signals silently no-op'd — making user cancellations indistinguishable from network failures in telemetry.

AbortSignal

• reason (read-only) and throwIfAborted().
• static AbortSignal.abort(reason?).
• aborted is now read-only (setter removed).
• Abort(reason) records the reason, defaulting to an AbortError — an Error whose name is "AbortError" (there is no DOMException polyfill, so this matches what web code checks: err.name === "AbortError").
• AbortController.abort(reason?) forwards the reason to the signal.

fetch

• Honors init.signal through the signal's JS interface (aborted / reason / add/removeEventListener), so fetch stays decoupled from the AbortController polyfill's C++ types.
• An already-aborted signal rejects the promise synchronously with the signal's reason, never touching the transport.
• Otherwise an "abort" listener cancels the in-flight transport via UrlRequest::Abort(), and the completion continuation rejects with the signal's reason (an AbortError) instead of a network-error TypeError. The listener is torn down when the request settles (breaking the listener ⟷ state ownership cycle).

Validation

Built and ran the UnitTests suite on Win32 / Chakra — all pass (199), including new tests:

• AbortSignal.abort() returns an aborted signal whose reason is an AbortError; throwIfAborted() throws only once aborted; abort(reason) records the reason; abort() defaults to an AbortError.
• fetch rejects with an AbortError both when the signal is already aborted and when aborted in-flight (the in-flight case cancels the WinRT transport and settles in ~50 ms).

Win32's UrlLib backend honors UrlRequest::Abort(), so the end-to-end abort path is exercised here.

Scope / follow-ups

• UrlRequest::Abort() on curl/NSURLSession: today it only cancels the Windows backend (the curl/NSURLSession backends never observe m_cancellationSource). Making abort interrupt those transports is a UrlLib change — until it lands, in-flight abort on Linux/Apple rejects only once the transport gives up on its own. Tracked in fetch/XHR transport failures and unhandled rejections lose all diagnostic fidelity before reaching the embedding app #196.
• AbortSignal.timeout() is left as a follow-up.

Part of the #196 chain (hops 2 and 3 are #200 and #201).

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[bkaradzic-microsoft]

Consolidated into #200 — the whole #196 chain is now a single PR (hop 4 is the last commit there). Closing to avoid three PRs for one feature in one repo, and because the AbortSignal/fetch changes share Fetch.cpp with hop 2 (now integrated cleanly in #200).